“…Some of the key features of QML and GMM estimators have been discussed in Pacurar (2008), Hautsch (2012), Brownlees et al (2012), Cipollini et al (2013), and Perera et al (2016), among others. Although these estimators have desirable asymptotic properties, their finite sample performance can at times be sensitive to the (unknown) conditional distribution of the observable process, and hence, in practice, when available, fully efficient maximum likelihood (ML) estimates are often preferred (see Grammig and Maurer, 2000;Perera and Silvapulle, 2021). For example, even though the asymptotic distribution of the QMLE is independent of the innovation distribution, when the data generating process is based on an innovation distribution that induces a non-monotonic hazard rate function (e.g.…”